llm-evaluation

Pass

Audited by Gen Agent Trust Hub on Jul 16, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is purely instructional, containing guidelines and Python code snippets for LLM quality evaluation. It does not perform network operations, access sensitive files, or execute external code.
  • [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that involves processing "real inputs from actual usage" to build evaluation sets. This creates a theoretical attack surface for indirect prompt injection where malicious data in the evaluation set could attempt to influence the LLM-as-judge. However, as the skill does not implement dangerous capabilities (such as shell execution or network requests) that could be triggered by such an injection, the risk is minimal.
  • Ingestion points: Evaluation sets built from "real usage" (SKILL.md).
  • Boundary markers: None explicitly provided in the JUDGE_PROMPT example.
  • Capability inventory: None; the provided code is limited to data analysis and score comparison.
  • Sanitization: None mentioned.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 16, 2026, 09:27 AM
Security Audit — agent-trust-hub — llm-evaluation